Table 2 Accuracy of the t-distribution-based P-value in estimating the probability of type I error in the frequency increment test
Sampling parametersα
NTL0.050.010.0010.0001
104101011.001.000.981.02
10410010101.001.001.001.04
1041,000101000.961.051.251.37
10310010100.991.081.311.38
1041,000101000.961.051.251.37
10510,000101,0000.730.680.650.62
1041005201.001.021.021.09
10410010101.001.001.001.04
10410010010.990.991.000.96
  • Columns 1–4 show simulation and sampling parameters (see text for notations). Columns 5–8 show the ratio of the true fraction of false positives in the FIT to the fraction α expected under the assumption that the frequency increment statistic is distributed according to Student’s t-distribution with L – 1 d.f., across a range of α-values. We performed 106 neutral Wright–Fisher simulations with the initial allele frequency ν0 = 0.5. Unlike our implementations of the LRT and the ELRT, the FIT can formally be applied in cases when the observed allele is either fixed or lost at the last sampled time point. Thus, the fraction of simulations in which absorption events prevented us from applying the FIT was <10−5.